AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Source Summaries

Principles for a new utopianism (Cave, DeepMind Institute, 2026)

medium confidence · updated 2026-09-20

DeepMind Institute launch essay by Cambridge philosopher Stephen Cave arguing that AGI discourse needs positive visions alongside risk management. Traces utopianism's twentieth-century discrediting, proposes three principles — medium-termism, humility and pluralism — intended to avoid the failures of earlier mass utopian movements, and offers universal basic services as a worked example, preferred over universal basic income.

Contents

"Principles for a new utopianism" is an essay of roughly 4,300 words with references and 38 footnotes, published on 16 September 2026 at the launch of the DeepMind Institute. Its author, Stephen Cave, is Professor and Academic Director of the Leverhulme Centre for the Future of Intelligence and Co-Director of the Institute for Technology and Humanity at the University of Cambridge, and was the only author among the Institute's inaugural pieces not employed by Google. The essay argues that risk management alone cannot orient AI policy, that utopian thinking is needed to supply direction, and that it can be rehabilitated if constrained by three principles. The Institute's disclaimer states that its pieces "should not be read as Google's official view."

Argument

The essay opens on an asymmetry: "it is easier to imagine the end of the world than the end of capitalism," and likewise easier than imagining "the end of poverty, privation, or war." Cave holds this to be especially true of Western depictions of AGI, where "apocalypse predominates," and connects it to polling showing that the share of Americans more concerned than excited about AI has "tipped into the majority." His objection to risk-centred framing is directional rather than dismissive: "Certainly AI does pose novel risks, and working to manage them is critical. But that alone does not make for a compelling vision of the future. It tells us what to steer away from, but not what to steer towards."

The historical section traces four centuries of proliferating utopias after Thomas More's 1516 coinage, through Lewis Mumford writing in 1922 that "our most important task at the present moment is to build castles in the air," to Mumford's own later recantation — "I was still living in the hopeful spirit of an earlier age" — after Auschwitz, the gulags and Hiroshima. Cave cites Russell Jacoby's observation that "most observers judge utopians or their sympathizers as foolhardy dreamers at best and murderous totalitarians at worst," and quotes Isaiah Berlin at length on the logic by which certainty about an ideal licenses any cost: "that was the faith of Lenin, of Trotsky, of Mao… of Pol Pot."

Against this he sets two developments. One is genuine improvement: extreme poverty falling continuously for two centuries, and child mortality before adulthood falling from half to 4%. The other he calls resignation — Herbert Marcuse's argument in The End of Utopia that "the abolition of hunger and of misery is possible with the productive forces that already exist," followed by Francis Fukuyama's 1989 end-of-history thesis, which Cave reads as producing "not so much a crowning end of history as a disillusioned end of hope."

Cave also rejects an identification of utopia with permanent happiness: humans "have not evolved to be perpetually happy," and a fully human life "will still be messy. Can you stub your toe in utopia? Yes." He defines utopia instead as "a society that offers the ideal conditions for human flourishing," with flourishing unpacked through health, wealth, happiness, meaning and social relationships.

The three principles

Medium-termism. Cave takes from Robert Nozick the image of "cavemen sitting together to think up what, for all time, will be the best possible society," and Nozick's question — "Do none of the reasons that make you smile at this apply to us?" Utopian projects should reach past crisis management, but should also accept that technological change will alter what seems possible and desirable. He offers three illustrations of values that AGI might shift rather than merely serve: meat-eating becoming unthinkable through better understanding of other species and better alternatives; current prison systems being looked back on "with horror and shame" after new crime-prevention and corrective measures; and lifelong marriage being seen as thraldom under much longer lifespans. He notes each has existing minority constituencies and none has the mainstream support that would let a government adopt it as a goal today. The principle also carries a practical objection to AGI utopianism that "leap[s] straight to positing a godlike superintelligence that can grant any wish," which he calls unhelpful both to decision-makers facing "institutional inertia, legacy infrastructure, material supply constraints, complex trade-offs, political resistance," and to the people already more concerned than excited, who "will be unmoved by scenarios that seem too sci-fi." He cites Ruth Levitas for treating utopia "as a method, not an end state."

Humility. One "can have a conception of how things might be better, without believing we must be entirely right. That path leads back to the gulag." The operative consequence Cave draws is a constraint on means: accepting that a vision may be wrong "should [make us] disinclined to compromise core values – such as fundamental human rights – to achieve it."

Pluralism. The worst twentieth-century atrocities in utopianism's name are ascribed to "faith in a singular vision of the optimal conditions for all people, regardless of their particular histories or preferences." A new utopianism should "reject these claims to universal validity, and embrace multiple ideas of the good life," letting individuals and communities act on their own priorities — "creating the change they wish to encounter," in Davina Cooper's phrase — and making coordinated change "as participatory as possible." Cave adds that the principles together allow utopian projects at different scales, some requiring states or intergovernmental cooperation and others realisable regionally or locally.

The worked example: universal basic services

Cave offers universal basic services as a project compatible with the three principles and contrasts it with universal basic income, which he identifies as "one currently popular utopian proposal in the face of AGI." His stated objection is that "putting cash in the hands of individuals – particularly in deprived areas – does not in itself do anything to improve the infrastructure on which flourishing is based," making UBI "a highly individualistic response to problems that are largely societal."

The alternative he sketches builds on the post-war welfare state: free public services of the highest quality — healthcare, education, public transport, communal spaces, museums, libraries. He notes UBS approximates what exists in some Nordic social democracies, which he describes as the happiest countries in the world, while arguing services could be better everywhere and would be transformative in many countries. AGI enters twice: raising prosperity enough to make such services affordable, and allowing each domain to be reimagined — free education extended to pre-school, universities and lifelong learning with more focus on individual interests and capabilities; healthcare made more personalised and preventive "with medical professionals freed to focus more on relational care." He names three further candidate areas without developing them: improving democratic processes, living more sustainably, and improving international architecture to reduce conflict.

Provenance and attribution

Fetched live on 17 September 2026 from institute.deepmind.com and converted from HTML. The launch and the essay's authorship are corroborated by Axios, "Google, DeepMind launch institute to explore AGI" (Maria Curi, 16 September 2026), and The Next Web, "Google DeepMind launches the DeepMind Institute to debate AGI" (Ana Maria Constantin, 16 September 2026), which lists the inaugural pieces with their authors. The page is dated to the public launch; its article:published_time metadata reads 4 September 2026, a content-management date earlier than the launch. A Firecrawl copy fetched the same day differs in a few standfirst and spelling details, indicating editing around launch; the capture follows the live HTML.

The acknowledgements thank sixteen people, among them Allan Dafoe and Shane Legg, both of Google DeepMind, Legg being the Institute's managing editor. The reference list includes Dario Amodei's Machines of Loving Grace (Machines of Loving Grace), Nick Bostrom's Deep Utopia, Berlin's The Crooked Timber of Humanity, Cooper's Everyday Utopias, Bregman's Utopia for Realists, and Birhane and colleagues on participatory AI.

Relationships

Categories: Source Summaries | ethics | global | opinion | workforce | academic
What links here (4)